Chicago Booth Review Podcast The ‘Professor of Uncertainty’ on AI
9 snips
Sep 17, 2025 Veronika Ročková, an econometrician at Chicago Booth, dives into the intriguing world of AI and its unpredictability. She discusses how AI's tendency to 'hallucinate' information can be harnessed creatively. By applying Bayesian methods, she shows how to leverage randomness for improved medical diagnoses and smarter galaxy classification. Veronika emphasizes the importance of using probabilistic uncertainty for better decision-making and shares tips on crafting effective prompts to enhance AI outputs across various fields.
AI Snips
Chapters
Transcript
Episode notes
Randomness Is A Feature Not A Bug
- Generative AI inherently produces randomness and should be treated as a distributional predictor, not a single-point answer.
- Embracing that randomness lets statisticians extract uncertainty measures that improve decision making.
Probe Models Repeatedly To Build Trust
- Ask the same AI question multiple times or rephrase prompts to assess consistency and build trust.
- Use repeated prompts to measure variability before relying on a model's output.
Use Bayesian Updating To Capture Uncertainty
- Bayesian inference updates prior beliefs with new data to produce posterior distributions that quantify uncertainty.
- Reporting full posterior distributions yields better decisions than relying on single-point estimates.